Bank-to-Turn Control for a Small UAV

using Backstepping and Parameter

Adaptation
Dongwon Jung and Panagiotis Tsiotras
Georgia Institute of Technology, Atlanta, GA 30332 USA
(e-mail: dongwon.jung, [email protected])
Abstract: In this research we consider the problem of path following control for a small xed-
wing unmanned aerial vehicle (UAV). Assuming the UAV is equipped with an autopilot for
low level control, we adopt a kinematic error model with respect to the moving Serret-Frenet
frame attached to a path for tracking controller design. A kinematic path following control law
that commands heading rate is presented. Backstepping is applied to derive the roll angle
command by taking into account the approximate closed-loop roll dynamics. A parameter
adaptation technique is employed to account for the inaccurate time constant of the closed-loop
roll dynamics during actual implementation. The path following control algorithm is validated in
real-time through a high-delity hardware-in-the-loop simulation (HILS) environment showing
the applicability of the algorithm on a real system.
1. INTRODUCTION
For decades unmanned aerial vehicles (UAVs) have been
proposed as replacements of human pilots during several
missions. In recent years small UAVs, in particular, are
increasingly receiving attention in universities, national re-
search institutes, etc., because of their envisioned missions
in both the military and civilian sectors.
In order to accomplish these missions with minimal human
intervention, the operation of a UAV should be fully
automated from the top level of path planning, down to
the inner control loop level. At the top level of the control
hierarchy, the path planning algorithm computes a rough
approximation of the optimal path toward the goal. A path
generation algorithm then smooths the path yielding a
dynamically feasible, yable path, by taking into account
the kinematic constraints of the UAV. Finally, the path
following algorithm is responsible for guiding the UAV to
stay close to the designed path.
Various control approaches in the literature have been
proposed to address the path following problem: Niculescu
(2001) introduced a lateral track control law, Park et al.
(2004) proposed a simple, yet eective, nonlinear control
logic and demonstrated it experimentally, Ren and Beard
(2004) considered the problem of constrained trajectory
tracking, Nelson et al. (2006) proposed a path following
control using vector elds to guide the UAV on the desired
path, and Rysdyk (2003) proposed a guidance law based
on good helmsman behavior.
Motivated by the method proposed by Micaelli and Sam-
son (1993), kinematic control laws have been used to regu-
late the distance error from the reference path for unicycle-
type mobile robots [Lapierre and Soetanto (2007); Lapierre
et al. (2007)]. The key aspect of the proposed algorithms is
that the control laws explicitly incorporate the controlled
motion of a virtual target along the path to be tracked.
In this article, we present a nonlinear path following algo-
rithm, which is an extension of the one by Lapierre and
Soetanto (2007) for UAV path following control. We apply
a standard backstepping technique to compute the roll
angle command from the heading rate command of the
kinematic control law. A parameter adaptation technique
is then applied to compensate for the inaccurate time con-
stant of the roll closed loop, yielding robust performance
during controller implementation. Besides rigorous proofs
for convergence of errors, the proposed path following algo-
rithm has an advantage of following a parametric reference
path with less stringent initial conditions, as compared to
the path following algorithms in the literature. The path
following control algorithm is validated through a high-
delity hardware-in-the-loop simulation (HILS) environ-
ment to show the applicability of the presented algorithm
on the actual system.
2. PROBLEM DESCRIPTION
A xed-wing UAV is equipped with a low-level autopilot
with on-board sensors that provides feedback control for
attitude angles, air speed, and altitude. In a typical search
mission, the air speed and the altitude are kept constant,
so that the UAV stays inside the safe ying envelope.
Suppose that the inertial speed (V ) and the course angle
(; inertial speed heading) are directly measured using an
on-board GPS sensor. A simplied kinematic model in the
two dimensional plane is utilized as follows,
x = V cos ,
y = V sin ,
= ,
(1)
where (x, y) denotes the inertial position, and is the
heading rate of the UAV. By expressing the equations of
motion in terms of the ground speed and the course angle,
the equations of motion become independent of the wind
velocity. Furthermore, by using inertial measurements for
path planning control, wind disturbance is naturally re-
jected so that the performance of path following can be
improved signicantly. For a xed-wing UAV at a banked-
turn maneuver with no sideslip, and assuming that || <

max
< /2, the heading rate is induced by the roll angle
with the inertial speed V as follows,
=
g
V
tan , (2)
I
Y
X
s
p
q
r

f
F
s
e
d
e
s
Fig. 1. Denition of the Serret-Frenet frame for the path
following problem.
where the roll angle is assumed to be controlled by an inner
PID loop. Properly tuned PID gains result in closed-loop
roll dynamics which resemble a rst-order system,

=
1

(
c
), (3)
where
c
is the roll angle command and

> 0 is the time

constant.
Consider now a UAV ying along a planar path as shown
in Fig. 1. Denote the inertial position of the UAV by
p = [x y]
T
R
2
. The geometric path is dened in terms
of the arc-length parameter s. For any given s, the inertial
position of the point associated with s is denoted by
q(s) R
2
, at which the Serret-Frenet frame is attached,
and moves along the path with speed s. The x-axis of the
Serret-Frenet frame is aligned with the tangent vector to
the path at q(s) and has an angle
f
(s) with respect to
the inertial frame I. Let the error vector e in the Serret-
Frenet frame be decomposed in the along-track error e
s
and the cross-track error e
d
. Then the inertial error vector
r = p q(s) expressed in the Serret-Frenet frame is
obtained by
e = R
T
(
f
)
_
p q(s)
_
, (4)
where,
R(
f
) =
_
cos
f
sin
f
sin
f
cos
f
_
, (5)
is the rotation matrix from the Serret-Frenet frame to the
inertial frame.
By dierentiating Eq. (4) with respect to time, it follows
that
e =

R
T
(
f
)
_
p q(s)
_
+R
T
(
f
)
_
p q(s)
_
, (6)
where,

R(
f
) = R(
f
)S(
f
), (7)
where, S(
f
) is the skew-symmetric matrix,
S(
f
) =
_
0
f

f
0
_
. (8)
It follows from Eq. (1) that
p = R()
_
V
0
_
. (9)
Notice that q is the time derivative of the point q(s), whose
speed is represented by s when expressed in the Serret-
Frenet frame. It follows that
q = R(
f
)
_
s
0
_
. (10)
Subsequently, by substituting Eqs. (9) and (10) in Eq. (6),
we obtain,
e = S(
f
)e +R(
f
)
_
V
0
_

_
s
0
_
, (11)
where R
T
(
f
)R() = R(
f
).
Let now the error course angle be dened by

f
. (12)
Hence,

=
f
=
f
, (13)
and

f

d
f
dt
=
d
f
ds
ds
dt
= (s) s, (14)
where (s) is the curvature of the path at q(s).
The error kinematic model of a xed-wing UAV for the
path following problem with respect to the Serret-Frenet
frame is summarized as follows,
e
s
= V cos (1 (s)e
d
) s, (15a)
e
d
= V sin (s)e
s
s, (15b)

= (s) s. (15c)
From Eqs. (15), the path following problem reduces into
a problem of driving the errors to zero as the UAV ap-
proaches the given path. In Micaelli and Samson (1993),
the point q is found by projecting p on the path, as-
suming the projection is well dened. Hence, the control
law derived by Micaelli and Samson (1993) requires a
stringent restriction on the initial position of q in order
to avoid singularities. In contrast, Lapierre and Soetanto
(2007) proposed to employ a moving Serret-Frenet frame
along the path, which eectively provides an extra con-
trol parameter s allowing q(s) to evolve along the error
states. This control parameter then mitigates the stringent
restriction on the initial condition arose in Micaelli and
Samson (1993).
3. PATH FOLLOWING CONTROLLER DESIGN
In this section we present a nonlinear path following
control logic, which steers the UAV to the reference path
with an inaccurately known time constant

. Beginning
with the derivation of a kinematic control law for the
heading rate command, we apply backstepping to derive
a roll control command for a xed-wing UAV, which, in
turn, induces an equivalent control eort to the kinematic
control law. In addition, we apply a parameter adaptation
technique to deal with an inaccurately known

.
3.1 Kinematic controller design
Following an approach similar to Lapierre and Soetanto
(2007) we rst derive a kinematic control law for the head-
ing rate. We introduce a bounded dierentiable function
with respect to the cross-track error e
d
, as follows
(e
d
)

e
2ke
d
1
e
2ke
d
+ 1
(16)
where, k > 0 and

(0, /2) [Micaelli and Samson

(1993)]. This function is so-called the approach angle, since
it provides the desired relative course transition of the
UAV to the path as a function of the cross-track error e
d
.
When the cross-track error e
d
is zero, the approach angle
vanishes, thus imposing the condition that the course angle
of the UAV must be tangential to the path. The positive
constant k sets the eectiveness of the transient maneuver
during approach. The approach angle provides a behavior
similar to that of a good helmsman [Rysdyk (2003)] since
it satises the following condition:
e
d
sin
_
(e
d
)
_
0, for all e
d
. (17)
A
B
Fig. 2. Local vector elds generated with respect to two
distinct points on the path.
This condition steers the UAV to the path along the
correct direction (turn left when the UAV is on the right
side of the path, and turn right in the opposite situation).
The kinematic controller for path following using the error
kinematic model in Eq. (15) is given below.
Proposition 1. [Lapierre et al. (2007)] Consider the kine-
matic error model of the UAV described in Eq. (15) and
the approach angle (e
d
) dened in Eq. (16). Assume that
the speed of the UAV is non-negative. Suppose that the
reference path is parameterized by the arc-length s, and
at each s the variables , e
s
, e
d
, and
f
are well dened.
Then, the following kinematic control law,
= k

_
(e
d
)
_
+(s) s +

(e
d
)
_
V sin (s)e
s
s
_

e
d
V

_
sin sin
_
(e
d
)
_
(e
d
)
_
, (18a)
s = k
s
e
s
+V cos , (18b)
where k
s
, k

and are positive constants, asymptotically

drives e
s
, e
d
, and towards zero.
Remarks. The objective of the closed loop system is to
steer the error states towards zero, so that the UAV follows
the virtual target moving on the path. It should be noted
that the approach angle plays an important role in steering
the UAV to the reference path. By constructing a vector
eld with respect to a reference point on the path the UAV
is guided to the path, as shown in Fig. 2. As mentioned
earlier, the moving reference frame in conjunction with
the extra control command s helps the errors converge
to zero. In particular, the term k
s
e
s
in Eq. (18) ensures
the convergence of e
s
to zero, providing the momentum
induced by the virtual target to reduce e
s
. Figure 2
shows the vector elds with respect to the corresponding
reference points on the path, which takes into account
the relative momentum induced by the virtual target
(drawn by dashed arrows). During implementation, the
new location of the virtual target on the path is propagated
using s obtained via a numerical integration scheme, such
as the forward Euler method. Hence, if the reference path
is well dened in terms of the arc-length parameter s, then
the corresponding variables e
s
, e
d
,
f
, and as also well
dened, and thus the proposed path following algorithm
is well-posed.
3.2 Roll angle command via backstepping
For a xed-wing UAV ying at a constant altitude, a
roll angle command is often used for heading control. A
simple way is to compute the roll angle command directly
from the heading rate command
d
using the kinematic
relationship in Eq. (2),

d
= tan
1
_
V
d
g
_
. (19)
Note that the actual response of the roll angle diers
from the commanded value of Eq. (19), but it can be
approximated by a rst-order system described as in
Eq. (3). Taking into account the approximated model of
Eq. (3), one can design a steering logic for the roll angle
command
c
in order to achieve
d
through
d
.
Following the standard backstepping technique [Spooner
et al. (2002)], we introduce an auxiliary control input for
the roll angle by augmenting the error model in Eq. (15)
with

= . Thus, the system is given as follows,
e
s
= V cos (1 (s)e
d
) s, (20a)
e
d
= V sin (s)e
s
s, (20b)

=
g
V
tan (s) s, (20c)

= , (20d)
where the auxiliary control input is associated with the
roll command by
=
1

_
. (21)
Suppose that the desired heading rate
d
is obtained from
Eq. (18a) and let
e

d
= (g/V ) tan
d
be an
error state for the heading rate, whose time derivative is
calculated using (tan )

= sec
2
, as follows

e
=
g
V

sec
2

d
. (22)
Proposition 2. Consider the kinematic error model of the
UAV described in Eqs. (20) and the approach angle
(e
d
) dened in Eq. (16). Assume that the speed of the
UAV is non-negative. Suppose that the reference path
is parameterized by the arc-length s, and at each s the
variables , e
s
, e
d
, and
f
are well dened. Then, the
control input,
=
V
g sec
2

_
k
e

e
+(e
d
) +
d
_
, (23)
where k
e
is a positive constant and
d
is given in Eq. (18a),
asymptotically drives e
s
, e
d
, and towards zero, while

d
. It should be noted that
d
can be calculated by
numerical dierentiation of
d
. However, in order to avoid
any high frequency noise eect due to dierentiation,
d
in
the actual implementation is obtained by using a pseudo
dierentiation lter in conjunction with saturation, as a
transfer function D(s) = sat
_
s/(s + 1)
_
where > 0 is
employed.
Proof. Let V
1
be a candidate Lyapunov function given
by,
V
1

1
2
_
e
2
s
+e
2
d
_
+
1
2
_
(e
d
)
_
2
+
1
2

2
e
. (24)
Dierentiating with respect to time one obtains,

V
1
=
1

_
es es + e
d
e
d
_
+
_
(e
d
)
__

(e
d
)
_
+ e e
=
1

es(V cos s) +
1

e
d
V sin
_
(e
d
)
_
+
_
(e
d
)
_
_
g
V
tan +
d

d
(s) s

(e
d
)
_
V sin (s)es s
_
+
e
d
V

sin sin
_
(e
d
)
_
(e
d
)
_
+ e
_
g
V

sec
2

d
_
.
(25)
By the denition of
e
, and substituting the desired
d
from Eq. (18a) and s from Eq. (18b), the term inside of
the bracket collapses to

V
1
=
k
s

e
2
s
+
e
d
V

sin
_
(e
d
)
_
k

_
(e
d
)
_
2
+
e
_
g
V

sec
2
+ (e
d
)
d
_
.
(26)
Choosing

through the auxiliary control input in Eq. (23),
that is,

= =
V
g sec
2

_
k
e

e
+(e
d
) +
d
_
, (27)
results in,

V
1
=
ks

e
2
s
+
e
d
V

sin
_
(e
d
)
_
k
_
(e
d
)
_
2
ke
2
e
0.
(28)
The last inequality implies, in particular, that e
s
, e
d
,
and
e
are bounded. Furthermore, (e
d
) is also bounded
from (16) since e
d
is bounded. It is also easy to show,
using (18a), that
d
is bounded. To this end, notice that
all signals in the rhs of (18a) except the last are bounded.
Moreover, since
sin sin
_
(e
d
)
_

_
(e
d
)
_
= sinc
_
1
2
_
(e
d
)
_
_
cos
_
1
2
_
+(e
d
)
_
_ (29)
and since the sinc function is bounded, the last term in
the rhs of (18a) is also bounded. Since
d
and
e
are
bounded, it follows that || < /2 and the control (23)
is well dened.
To complete the proof, note that V
1
is radially unbounded,
hence the set
c
= {V
1
(e
s
, e
d
, ,
e
) c} is a compact,
positively invariant set. Let E
1
be the set of all points in

c
where

V
1
= 0. The set E
1
is given by E
1
= {e
s
=
e
d
= 0, = , and (g/V ) tan =
d
}. Noticing that ()
is a function of the cross-track error e
d
, one can easily
verify that any point starting from the set E
1
will remain
in the set, i.e. E
1
is an invariant set. By the LaSalles
invariance principle, we have that every trajectory starting
inside the set
c
approaches E
1
as t . Therefore,
lim
t
e
s
= 0, lim
t
e
d
= 0, and lim
t
= 0 since
(e
d
) 0. We also have (g/V ) tan
d
as t .
3.3 Parameter adaptation
Note that the actual roll angle command is computed from
Eq. (3) and (23),

c
=

+, (30)
where,

is the time constant of the system.

In practice,

is determined by the characteristics of

the UAV airframe and PID gains, thus it is not known
accurately. The actual roll command

c
is then given by

c
=

+, (31)
where

is an estimate of

.
In order to compensate for any uncertainty in

, we apply
a parameter adaptation technique. To this end, let V
2
be
a candidate Lyapunov function dened by
V
2
V
1
+
1
2
1
k
a

, (32)
where V
1
is given in Eq. (24) and

is the
estimate error. Dierentiating with respect to time, we get

V
2
=

V
1
+
1
ka

=
ks

e
2
s
+
e
d
V

sin
_
(e
d
)
_
k
_
(e
d
)
_
2
+ e
_
g
V

sec
2
+ (e
d
)
d
_
+
1
ka

.
(33)
The actual value of

is calculated from Eq. (3) in
conjunction with the actual roll command in Eq. (31) and
given in Eq. (23). Subsequently, substituting the actual

in Eq. (33), we get

V
2
=
ks

e
2
s
+
e
d
V

sin
_
(e
d
)
_
k
_
(e
d
)
_
2

ke
2
e
+ e

_
(e
d
)
d
_
+
1
ka

.
(34)
Choosing

as,

= k
a

e
_
(e
d
)
d
_
, (35)
where k
a
is a positive constant. It follows that

V
2
=
k
s

e
2
s
+
e
d
V

sin
_
(e
d
)
_
k

_
(e
d
)
_
2

k
e

2
e
0.
(36)
Assuming

is constant, the parameter update law is

readily obtained from Eq. (35) as follows,

= k
a

e
_
(e
d
)
d
_
. (37)
Proposition 3. Let the control law in Eqs. (18) and (23).
With the parameter update law given by Eq. (37) the
actual roll command of Eq. (31) guarantees that the
signals e
s
, e
d
, and asymptotically tend to zero, while
(g/V ) tan
d
.
Proof. In order to prove the proposition notice that
lim
t
V
2
(t) exists since V
2
is bounded from below and
is non-increasing. It therefore suces to show that

V
2
is uniformly continuous, in which case we can invoke
Barbalats lemma to ensure that lim
t

V
2
(t) = 0.
To this end, consider the expression of

V
2
in Eq. (36),
from which it follows that e
d
, e
s
, ,
e
,

, and

are
all bounded. As with the proof of Proposition 2, it follows
from (18b) that s is bounded, and from (18a) that
d
is
bounded as well. The boundedness of
d
and
e
imply
that || < /2. Using (20) we conclude that e
s
, e
d
,

are
also bounded. Dierentiating Eq. (18a) with respect to
time, one can show that all terms are bounded, while the
last term is also bounded since the derivative of the sinc
function is bounded (see also Eq. (29)). Hence, it can be
shown that
d
is bounded. Furthermore, (e
d
) and its time
derivative

(e
d
) =

(e
d
) e
d
are bounded. A straightforward
calculation shows that

e
=

_
k
e

e
+(e
d
)
_

d
. (38)
From the previous equation it follows that
e
is bounded.
Furthermore,

is also bounded from (37). Consider now

the expression for

V
2
, given by

V
2
=
ks

es es +
V e
d

sin
_
(e
d
)
_
+
V e
d

(e
d
) cos
_
(e
d
)
_
2k
_
(e
d
)
__

(e
d
)
_

kee e,
(39)
Table 1. Simulation parameters.
V = 20 [m/s] k = 0.01 ks = 0.4 k = 0.001 |
max
c
| = /6
ke = 1.1 ka = 0.7 = 4000

1.1
where, all expressions in the rhs of the equation have
been shown to be bounded. It follows that

V
2
is bounded
and hence

V
2
is uniformly continuous. Applying Barbalats
lemma, it follows that

V
2
0 as t 0.
4. HARDWARE IN-THE-LOOP SIMULATION
RESULTS
4.1 Simulation environment
A realistic hardware-in-the-loop simulation (HILS) envi-
ronment has been developed to validate the UAV autopi-
lot hardware and software development utilizing Matlab

and Simulink

. A full 6-DOF nonlinear aircraft model is

used in conjunction with a linear approximation of the
aerodynamic forces and moments, along with Earth grav-
itational (WGS-84) and magnetic eld models. Detailed
models for the sensors and actuators have also been incor-
porated. Four independent computer systems are used in
the hardware-in-the-loop simulation (HILS) as illustrated
in Fig. 3. A 6-DOF simulator, the ight visualization
computer, the autopilot micro-controller, and the ground
station computer console are involved in the simulation.
Further details about the UAV platform, autopilot and
HILS set-up can be found in Jung et al. (2005); Jung and
Tsiotras (2007a) and Jung and Tsiotras (2007b).
Computer
Wireless RF Modem
HIL Bridge
(RS232)
Binary
States
Control
autopilot
Visualization
(UDP)
Flight
Dynamics
Simulator
Flight
Dynamics
Visualization
RC to
USB
Adpator
900MHz
Wireless
Binary
G
ro
und
Station
;FlightGear v1.9
;Cockpit view
;Matlab/Simulink
;6DOF nonlinear model
;Real -time simulation
;Remote pilot switching
;Flight control executable
- Inner/Outer loop controller
;Sensor data processing (20Hz)
;Communication to GS
;Ground station GUI
;Communication to autopilot
;Data logging / monitoring
;High-level controller
;RS232 Simulink library
;Real-time Simulink execution
;Bi-directional communication
Fig. 3. High delity hardware-in-the-loop simulation
(HILS) environment for validating the path following
algorithm
4.2 Simulation results
This section illustrates the performance of the derived
path-following control law. The reference path is given by
a quartic B-spline over a non-decreasing knot parameter
u which is monotonic to the arc-length s. Hence, we
can compute q(s),
f
(s), and (s) in terms of the knot
parameter as follows,

f
(s) = tan
1
y

, (s) =
x

(x
2
+y
2
)
3/2
. (40)
where ()

and ()

are the derivatives with respect to u.

The knot parameter is propagated along with Eq. (18) as
follows,
du
dt
=
ds
dt
_
ds
du
= s/
_
x
2
+y
2
. (41)
The parameters used in the simulations are shown in
Table 1. We present the results from two simulations.
In the rst case, we calculate the roll angle command
0 100 200 300 400 500 600 700 800 900 1000
0
100
200
300
400
500
600
700
800
East [meter]
N
o
r
t
h

[
m
e
t
e
r
]

: Ref. path
: Position
Fig. 4. Reference path and actual trajectory of the UAV
without parameter adaptation.
0 10 20 30 40 50 60 70 80
140
120
100
80
60
40
20
0
20
40
Time [sec]
E
r
r
o
r

s
t
a
t
e
s

[
m
e
t
e
r
]

: e
s
: e
d
(a) Track errors
0 10 20 30 40 50 60 70 80
40
20
0
20
40
60
80
Time [sec]
E
r
r
o
r

c
o
u
r
s
e

a
n
g
l
e

[
d
e
g
]
(b) Course angle error
0 10 20 30 40 50 60 70 80
20
10
0
10
20
30
40
50
Time [sec] K
i
n
e
m
a
t
i
c

c
o
n
t
r
o
l
l
e
r

o
u
t
p
u
t
,

d

[
d
e
g
/
s
e
c
]
(c) Heading rate command
0 10 20 30 40 50 60 70 80
40
30
20
10
0
10
20
30
40
Time [sec]
R
o
l
l

a
n
g
l
e

a
n
d

c
o
m
m
a
n
d

[
d
e
g
]

:
c
:
(d) v/s c
Fig. 5. Error states and command inputs without param-
eter adaptation.
from Eq. (30) using a time constant

= 0.4 which is
inaccurately known. Without parameter adaptation, as
shown in Fig. 4, we observe a sluggish and low damped
response of the actual trajectory. The error variables are
shown in Fig. 5. In the second case, we calculate the roll
angle command from Eq. (31) with the parameter update
law of Eq. (37). Figures 6-8 show the results. The error
states tend to zero asymptotically as shown in Fig. 7.
Because the roll angle command is limited within /6,
the UAV is unable to exactly follow the path where the
curvature exceeds the maximum curvature achievable by
the UAV at the speed of 20 [m/sec]. Nevertheless, the
path following control law forces the UAV converge to the
path asymptotically after a short transient. It should be
noted that the simulation was carried out in conjunction
with realistic sensor noises and constant wind disturbance,
showing that the adaptation law is more or less robust to
the presence of such measurement noises and disturbances.
Fig. 8 displays the time history of the estimate of

,
which shows that from the initial guess of

= 0.4,
the parameter estimate converges to the actual value of

1.1.
0 100 200 300 400 500 600 700 800 900
0
100
200
300
400
500
600
700
800
East [meter]
N
o
r
t
h

[
m
e
t
e
r
]

: Ref. path
: Position
Fig. 6. Reference path and actual trajectory of the UAV
with parameter adaptation.
0 10 20 30 40 50 60 70 80
140
120
100
80
60
40
20
0
20
40
Time [sec]
E
r
r
o
r

s
t
a
t
e
s

[
m
e
t
e
r
]

: e
s
: e
d
(a) Track errors
0 10 20 30 40 50 60 70 80
40
20
0
20
40
60
80
Time [sec]
E
r
r
o
r

c
o
u
r
s
e

a
n
g
l
e

[
d
e
g
]
(b) Course angle error
0 10 20 30 40 50 60 70 80
20
10
0
10
20
30
40
50
Time [sec] K
i
n
e
m
a
t
i
c

c
o
n
t
r
o
l
l
e
r

o
u
t
p
u
t
,

d

[
d
e
g
/
s
e
c
]
(c) Heading rate command
0 10 20 30 40 50 60 70 80
40
30
20
10
0
10
20
30
40
Time [sec]
R
o
l
l

a
n
g
l
e

a
n
d

c
o
m
m
a
n
d

[
d
e
g
]

:
c
:
(d) v/s c
Fig. 7. Error states and command inputs with parameter
adaptation.
0 10 20 30 40 50 60 70 80
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
Time [sec]
E
s
t
i
m
a
t
e
d

p
a
r
a
m
t
e
r
,

Fig. 8. Parameter estimate of

.
5. CONCLUSION
In this research a nonlinear path following control law has
been developed for a small UAV using backstepping of the
heading rate command. The kinematic control law realizes
cooperative path following so that the motion of a virtual
target is controlled by an extra control input to help
the convergence of the error variables. A roll command
that gives rise to the desired heading rate has been
derived by taking into account the inaccurate system time
constant. The path following control algorithm is validated
through a high-delity hardware-in-the-loop simulation
(HILS) environment, which veries the applicability of the
presented algorithm to the actual UAV.
ACKNOWLEDGEMENTS
Partial support for this work has been provided by NSF
award CMS-0510259.
REFERENCES
D. Jung, E. Levy, D. Zhou, R. Fink, J. Moshe, A. Earl,
and P. Tsiotras. Design and Development of a Low-
Cost Test-Bed for Undergraduate Education in UAVs.
In Proceedings of the 44
th
IEEE Conference on Decision
and Control, pages 27392744, Seville, Spain, December
2005.
D. Jung and P. Tsiotras. Inertial Attitude and Position
Reference System Development for a Small UAV. In
AIAA Infotech at Aerospace, Rohnert Park, CA, May
2007a. AIAA Paper 07-2768.
D. Jung and P. Tsiotras. Modelling and Hardware-in-the-
loop Simulation for a Small Unmanned Aerial Vehicle.
In AIAA Infotech at Aerospace, Rohnert Park, CA, May
2007b. AIAA 07-2763.
L. Lapierre and D. Soetanto. Nonlinear Path-following
Control of an AUV. Ocean Engineering, 34:17341744,
2007.
L. Lapierre, R. Zapata, and P. Lepinay. Combined Path-
following and Obstacle Avoidance Control of a Wheeled
Robot. The International Journal of Robotics Research,
26(4):361375, April 2007.
A. Micaelli and C. Samson. Trajectory Tracking
for Unicycle-type and Two-Steering-Wheels Mobile
Robots. Technical Report 2097, INRIA, Sophia-
Antipolis, November 1993.
D. R. Nelson, D. B. Barber, T. W. McLain, and R. W.
Beard. Vector Field Path Following for Small Unmanned
Air Vehicles. In Proceedings of the 2006 American
Control Conference, pages 57885794, Minneapolis, MN,
June 2006.
M. Niculescu. Lateral Track Control Law for Aerosonde
UAV. In 39
th
AIAA Aerospace Sciences Meeting and
Exhibit, Reno, NV, January 2001. A01-16013.
S. Park, J. Deyst, and J. P. How. A New Nonlinear
Guidance Logic for Trajectory Tracking. In AIAA Guid-
ance, Navigation, and Control Conference and Exhibit,
Providence, RI, August 2004. AIAA-2004-4900.
W. Ren and R. W. Beard. Trajectory Tracking for
Unmanned Air Vehicles with Velocity and Heading Rate
Constraints. IEEE Transactions on Control Systems
Technology, 12(5):706716, September 2004.
R. Rysdyk. UAV Path Following for Constant Line-Of-
Sight. In 2nd AIAA Unmanned Unlimited Conference
and Workshop and Exhibit, San Diego, CA, 2003. AIAA-
2003-6626.
J. T. Spooner, M. Maggiore, R. Ord o nez, and K. M.
Passino. Stable Adaptive Control and Estimation for
Nonlinear Systems. A John Wiley & Sons, Inc., New
York, NY, 2002.